The Trump administration asked OpenAI not to release GPT-5.6 directly to the general public over security concerns. Sam Altman’s company will first provide the model to a limited number of customers, Reuters and The Information reported.
According to the reports, the request came from the White House Office of the National Cyber Director and the Office of Science and Technology Policy. The agencies asked to limit the GPT-5.6 launch while the administration develops an approach to testing and evaluating the safety of new AI models.
OpenAI has not publicly commented on the restricted release, but The Information said Altman told employees about it. A source familiar with the matter told Reuters that authorities asked for a phased rollout of GPT-5.6.
How the release will work
According to The Verge, the model will go to a small group of enterprise customers. The outlet reported that the federal government will decide which clients can access the model during the preview phase.
“We have made it clear to the U.S. government that this is not our preferred long-term model, and we will work with it and other industry participants on a more sustainable approach to future releases,” Altman said in a message to employees, according to The Information.
Axios, citing a source, reported that OpenAI had worked with the administration in advance on the GPT-5.6 release. According to the source, the White House was familiar with the new model’s capabilities and could assess them in advance. The outlet also said that on June 25 Altman discussed the matter with U.S. Commerce Secretary Howard Lutnick. He allegedly wanted to ensure that the relevant government bodies had tested and approved the model.
Why the government intervened
Sources linked the request to the new model’s capabilities in sensitive areas. According to a CNN source, OpenAI and the administration view GPT-5.6 as comparable to Mythos. An unnamed expert cited by Axios made a similar comparison, adding that authorities want to ensure sufficient safeguards in tools of this class.
A White House spokesperson told CNN the administration continues to work with developers of frontier AI models to establish common approaches to scaling challenges.
The request to OpenAI coincided with a broader overhaul of U.S. policy on frontier artificial intelligence. On June 2, Trump signed an order titled Promoting Advanced Artificial Intelligence Innovation and Security. The document does not mandate licensing of AI model releases, but instructs several agencies to develop a classified benchmarking process to assess their advanced cyber capabilities.
The order also provides a voluntary mechanism for engagement with developers. Companies can provide the government with frontier models for up to 30 days before release. For trusted organizations, such access should be accompanied by requirements for confidentiality, cybersecurity, insider-risk protections, and intellectual property protection.
Contrast with Anthropic
OpenAI’s situation appears softer than Anthropic’s recent case. On June 9, the startup released two versions of the same Claude model family. The company described Fable 5 as a Mythos-class system but safe for general use. Claude Mythos 5 is a “private” base model with relaxed constraints in certain areas.
However, on June 12 Anthropic disabled the models due to a U.S. government directive under export controls. The firm said the document formally bans access for any foreign nationals, including foreign employees. Later, media linked the restrictions to China, and some sources cited reports of bypassing Fable 5’s safeguards as another trigger.
Against this backdrop, market participants and experts pointed to a lack of a transparent process. CNN quoted Public First head Brad Carson, who called the Fable situation an example of the need for clear rules.
“The Fable episode shows the need for clear regulation. Right now you have a special, personalized, opaque, and possibly unlawful approach,” he said.
Carson added that the government can intervene in the case of dangerous products, including AI models, but such a process should adhere to principles of basic fairness.
OpenAI’s stance on regulation
In early June, OpenAI presented a framework for governing frontier AI. The company proposed strengthening the Artificial Intelligence Standards and Innovation Hub and building a broader plan to bolster government resilience to risks from next-generation models.
“The federal government should now build on this foundation a durable federal framework that can evolve alongside the technology,” the document says.
In a separate document, OpenAI described its approach to assessing and mitigating risks in areas of cyber offense, CBRN threats, harmful manipulation, loss of control, model reporting, safety risk management, incident response, and external expert review.
Earlier, the company confidentially filed an S-1 registration statement with the U.S. Securities and Exchange Commission for a potential IPO. According to Reuters, OpenAI is preparing for a public listing at a $1 trillion valuation.
The Wall Street Journal, citing sources, wrote about a possible listing in September and the involvement of Goldman Sachs and Morgan Stanley in preparing the deal. In its new statement, OpenAI did not confirm those timelines. According to NYT, advisers offered management two scenarios: wait until 2027 to target a $1 trillion valuation or agree to an earlier listing at a lower level. Altman, Reuters wrote, rejected the second option.
On June 22, OpenAI launched the full version of GPT-5.5-Cyber — a specialized model for finding, verifying, and fixing vulnerabilities. The release came amid the restrictions around Anthropic.
